Hi all, We have written a blog article that compares Hive 4 on MR3 and Trino 453 (the latest release of Trino) using the 10TB TPC-DS benchmark. While it does not evaluate vanilla Hive, the results might be useful to those interested in the overall speed of Hive 4 in LLAP mode.
https://www.datamonad.com/post/2024-08-01-hive-4.0-performance-1.11/ Thanks, --- Sungwoo On Wed, Jul 31, 2024 at 12:00 AM David Engel <da...@istwok.net> wrote: > Congratulations, Sungwoo. I look forward to trying out Hive 4.0.0 on > MR3 when my time allows. > > David > > On Tue, Jul 30, 2024 at 11:29:02PM +0900, Sungwoo Park wrote: > > Hi all, > > > > We would like to announce the release of Hive 4.0.0 on MR3. It is based > on > > Hive 4.0.0 (together with 17 commits from hive-4.0.1-must) and replaces > Tez > > by a new execution engine MR3 (version 1.11). We have also released Hive > > 3.1.3 on MR3 which backports about 800 patches to branch-3.1. > > > > The git repo for Hive 4.0.0 on MR3 is available at: > > https://github.com/mr3project/hive-mr3/tree/master4.0 > > > > Compared with Apache Hive, Hive on MR3 (Hive-MR3) has several new > features. > > > > - Hive-MR3 is easy to set up because it automatically achieves the speed > of > > LLAP without requiring any extra steps. LLAP IO cache is also easy to > > enable. > > - Hive-MR3 can run directly on Kubernetes. On public clouds, you can take > > advantage of autoscaling and spot instances. We provide shell scripts, > helm > > chars, and TypeScript code for executing Hive-MR3 on Kubernetes. > > - Hive-MR3 can run in standalone mode (similarly to Spark standalone mode > > and Presto/Trino). > > - Hive-MR3 supports Java 17. > > - A single master manages all jobs while containers are shared by all > > running jobs. As a result, Hive-MR3 yields much higher throughput for > > concurrent batch jobs. > > > > All the documentation on Hive-MR3 is at: https://mr3docs.datamonad.com/ > > > > Thanks, > > > > --- Sungwoo > > -- > David Engel > da...@istwok.net >
